Introduction to Fourier Transform

Introduction

The Fourier Transform is arguably the single most important mathematical tool in all of signal processing and communications engineering. It provides a way to decompose any signal — not just periodic ones — into its constituent frequencies. While the Fourier series handles periodic signals by breaking them into harmonics, the Fourier transform handles aperiodic signals by breaking them into a continuous spectrum of frequencies. Together, they give us a complete frequency-domain picture of any signal we might encounter.

For B.Tech students, the Fourier transform is the gateway to understanding bandwidth, filtering, modulation, sampling, and spectral analysis. It transforms complex time-domain problems into often-simple frequency-domain problems. Master this tool, and you hold the key to most of signal processing.

Motivation: Why We Need the Fourier Transform

The Fourier series works beautifully for periodic signals, giving us a discrete set of harmonic frequencies. But what about:

• A single rectangular pulse (not repeating)?
• An exponential decay after a switch is flipped?
• A speech utterance (starts and stops)?
• A single musical note that fades away?

These are aperiodic signals — they don't repeat. They cannot be represented by a discrete set of harmonics. Instead, they have energy spread across a continuum of frequencies. The Fourier transform captures this continuous spectral distribution.

Conceptual Development

Think of an aperiodic signal as a periodic signal with period $T_0 \to \infty$. As the period grows:

1. The harmonic spacing $\Delta\omega = 2\pi/T_0$ gets smaller and smaller
2. The discrete spectral lines get closer together
3. In the limit, they merge into a continuous function
4. The sum $\sum c_n e^{jn\omega_0 t}$ becomes an integral $\int X(\omega)e^{j\omega t}d\omega$

This is not just a mathematical trick — it is the actual physical meaning. An aperiodic signal has infinitesimal contributions from every frequency, not finite contributions from discrete harmonics.

The Fourier Transform Pair

Analysis equation (time → frequency): $$X(\omega) = \int_{-\infty}^{\infty} x(t) e^{-j\omega t} dt$$

Synthesis equation (frequency → time): $$x(t) = \frac{1}{2\pi}\int_{-\infty}^{\infty} X(\omega) e^{j\omega t} d\omega$$

$X(\omega)$ is called the spectrum, spectral density, or frequency-domain representation of $x(t)$.

The analysis integral correlates the signal with each complex exponential $e^{j\omega t}$ — extracting how much of each frequency is present. The synthesis integral reconstructs the signal by summing all frequency contributions.

What the Fourier Transform Tells Us

For a complex-valued $X(\omega) = |X(\omega)|e^{j\theta(\omega)}$:

• $|X(\omega)|$ (magnitude spectrum): The amplitude density at frequency $\omega$ — tells you WHICH frequencies are present and HOW MUCH
• $\theta(\omega)$ (phase spectrum): The phase at each frequency — tells you the TIMING relationships between frequency components
• $|X(\omega)|^2$ (energy spectral density): Energy distribution across frequencies

Both magnitude AND phase are needed to reconstruct the signal. If you keep only the magnitude and discard phase, you lose the signal's shape (though the spectral energy content is preserved).

Conditions for Existence

The Fourier transform exists (converges) if:

Sufficient condition: $\int_{-\infty}^{\infty}|x(t)|dt < \infty$ (absolutely integrable)

This guarantees $X(\omega)$ is finite and continuous. However, many useful signals (step function, sinusoids, constants) don't satisfy this condition. Their transforms are defined using generalized functions (delta functions):

• $u(t) \leftrightarrow \pi\delta(\omega) + 1/(j\omega)$
• $\cos(\omega_0 t) \leftrightarrow \pi[\delta(\omega-\omega_0) + \delta(\omega+\omega_0)]$
• $1 \leftrightarrow 2\pi\delta(\omega)$

Key Transform Pairs to Remember

Frequency-Domain Thinking

Once you internalize the Fourier transform, you develop "spectral intuition":

• Narrow in time → wide in frequency (and vice versa): The uncertainty principle
• Smooth signals → rapidly decaying spectra: Discontinuities cause slow spectral decay
• Real signals → conjugate-symmetric spectra: $X(-\omega) = X^*(\omega)$
• Multiplication in time → convolution in frequency: Modulation spreads the spectrum
• Convolution in time → multiplication in frequency: Filtering shapes the spectrum

A Simple Example

Find the Fourier transform of a one-sided exponential: $x(t) = e^{-2t}u(t)$.

$$X(\omega) = \int_0^{\infty} e^{-2t}e^{-j\omega t}dt = \int_0^{\infty}e^{-(2+j\omega)t}dt = \left[\frac{e^{-(2+j\omega)t}}{-(2+j\omega)}\right]_0^{\infty} = \frac{1}{2+j\omega}$$

Magnitude: $|X(\omega)| = \frac{1}{\sqrt{4+\omega^2}}$ — a low-pass spectrum peaking at $\omega=0$ (DC) and rolling off at higher frequencies.

Phase: $\angle X(\omega) = -\arctan(\omega/2)$ — negative, indicating the signal is causal (delayed).

3-dB bandwidth: $|X|$ drops to $1/\sqrt{2}$ of its peak at $\omega = 2$ rad/s, so $B_{3dB} = 2$ rad/s = $1/\pi$ Hz. Notice: the bandwidth is inversely related to the decay rate. Faster decay ($a$ larger) → wider bandwidth.

Key Takeaways

• The Fourier transform extends Fourier series to aperiodic signals: discrete harmonics → continuous spectrum
• Analysis: $X(\omega) = \int x(t)e^{-j\omega t}dt$; Synthesis: $x(t) = \frac{1}{2\pi}\int X(\omega)e^{j\omega t}d\omega$
• Magnitude spectrum shows frequency content; phase spectrum shows timing; both are needed for reconstruction
• Key duality: narrow in time ↔ wide in frequency (uncertainty principle)
• Signals with discontinuities have slowly decaying spectra ($\sim 1/\omega$); smooth signals decay faster
• The convolution theorem ($Y = X \cdot H$) makes frequency-domain analysis of LTI systems trivial